AI Could Make Us More Productive and Just as Busy

Suppose a task that takes you an hour suddenly takes ten minutes. It seems reasonable to expect fifty minutes of free time. At work, though, that depends on whether anyone still wants the same amount of output. If your manager can now ask for six versions of a presentation instead of one, the afternoon fills up again. The technology has delivered exactly what it promised. You are producing more with the same effort. You just have not gained the thing you actually wanted.

Economists have a useful explanation for this. When something becomes cheaper, people tend to use more of it. Improvements in energy efficiency, for example, lower the cost of using an appliance and can encourage greater use. This is called the rebound effect. The stronger claim associated with the Jevons paradox is that demand increases enough to outweigh the original savings entirely. That outcome is possible, but efficiency does not automatically cause it. How much demand responds matters.

The same question applies to office work. Imagine a business that used to spend twenty hours preparing one sales proposal. With AI, it can prepare the same proposal in five. If it still submits one proposal, the saving is substantial. If the lower cost makes it worthwhile to pursue five potential clients, preparation now takes twenty five hours. That is an illustration of how demand could absorb a productivity gain, rather than a prediction about every business. Some companies will find plenty of additional worthwhile projects. Others will run out of customers.

This complicates the usual argument about automation and jobs. A tool can reduce the labor required for each task while increasing the number of tasks that businesses want completed. James Bessen’s research on automation emphasizes this role of demand. The effect on employment depends partly on whether customers keep buying more as production becomes cheaper. There is no guarantee that displaced workers get the new jobs, or that the extra work requires the same skills. An expanding market can still be painful for particular workers.

There is also a difference between growing demand and growing expectations. A client who once accepted an answer tomorrow may begin expecting it this afternoon. A report that previously needed three scenarios may suddenly need fifteen. Each request can look reasonable in isolation, especially when the software makes the first draft almost effortless. Someone still has to check the assumptions and take responsibility for the result. The easiest part of the job gets faster while the remaining work becomes more demanding.

I think this is why promises about AI and leisure deserve more scrutiny. The relevant question is what employers, customers and workers will do with the capacity it creates. Some of it will become new products and lower prices. Some may become higher profits or shorter hours. Those outcomes depend on competition, bargaining and what people choose to buy. Saving time on a task gives us an opportunity to work less. Whether we take it is an economic decision.